Is your feature request related to a problem? Please describe.
Some applications have restrictions on the total number of MPI ranks over which they can decompose their computations. For example, several of the NPB benchmarks require power-of-two or square-number rank counts. Depending on the node count and the experiment platform's count of cores per node, some nodes may contain fewer ranks than other nodes.

Right now, the appconf class only supports defining a count of ranks per node, which the experiment launcher multiplies against node count to determine the total rank count. This results in an incorrect total rank count. Although the application is able to execute in that case, GEOPM does not count epochs on nodes that do not check in to an epoch with every assigned rank.

Describe the solution you'd like
I would like for appconfs to be able to declare the total rank count needed for an application. Many apps are fine with the current behavior, so it would also be nice if it is possible for appconfs to omit such a declaration (defaulting to the current nodes*ranks_per_node behavior) so that we don't need to modify existing configs.